Reproduce “EgoCPT: Learning Dexterous Manipulation from a Single Ego-Centric Video with Contact-Aware Optimization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhixin Sun, Kairui Shi, Tianao Ren, Ming Yang, Mark Cutkosky, James Jingxi Xu
EgoCPT reconstructs robot-feasible contact, force, and motion references from human demonstrations before policy learning. It preserves demonstrated object motion while searching for robot-side contacts and forces, then uses a residual reinforcement-learning policy to track the objects. In simulation, it achieves 67.3% success on 52 noisy EgoDex sequences and 92.0% on 25 TACO sequences, under stated object-motion thresholds.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.